53 research outputs found
On the automation of high level synthesis of convolutional neural networks
No full textConvolutional Neural Networks (CNNs) are a particular type of Artificial Neural Networks (ANNs) inspired by cells in the primary visual cortex of animals, and represent the state of the art in image recognition and classification. Nowadays, such supervised learning technique is very popular in Big Data analytics. In this context, due to the huge amount of data to be processed, it is crucial to find techniques to speed up the computation. In particular, the dataflow pattern of CNN algorithm results to be suitable for hardware acceleration. This paper proposes a framework to automatically generate a hardware implementation of CNNs on Field Programmable Gate Arrays (FPGAs), based on the High Level Synthesis (HLS) of configurable offline-trained networks
Interstitial Lung Disease after Kidney Transplantation and the Role of Everolimus
No full textBackground Kidney transplant recipients are at higher risk of developing pulmonary complications related to immunosuppression, and inhibitor of the mammalian target of rapamycin (mTORi) has been reported as a potential cause. Methods Five hundred kidney-transplanted patients were retrospectively analyzed for pulmonary complications on the basis of clinical and instrumental data (chest radiography, high-resolution computed tomography, broncho-alveolar lavage, oximetry). Results We found 26 interstitial lung diseases (ILD) (16%): 12 cases (46.2%) were from infections (42.8% by Pneumocystis jirovecii) and 14 cases of ILD (53.8%) resulted as drug-induced ILD (DI-ILD). According to anti-rejection protocols, DI-ILD occurred in 8 patients (57%) while on triple regimen including steroids, everolimus (EVL), and cyclosporine (CyA) and in 6 patients on double regimen with steroids and mTORi: EVL or sirolimus (43%). In ILD+ patients, everolimus trough-concentration (EVLTLC) and cyclosporine (2nd-hour concentration: CyAC2) levels were higher than in patients in the same regimen but with ILD- (EVLTLC [ng/mL] 9.84 versus 6.85; CyAC2 [ng/mL] 303.97 versus 298.56). The formula that used the combined blood levels of both drugs (EVLTLC + CyAC2/100) resulted in a significant difference between groups of patients (12.88 ± 1.61 versus 9.83 ± 1.91). Applying receiver operator characteristic curve (ROC) analysis to detect risk of developing ILD when on combined protocol with EVL and CyA, we obtained an area under the curve of 0.8622 (P =.0081) and 0.9082 (P =.0028), respectively, when using EVLTLC or the combination formula with both drugs. Conclusions In renal transplant patients, we obtained a relationship of ILD to specific drug concentration. On the basis of ROC analysis, patients on EVL and CyA combined protocol are at risk of ILD when EVLTLC is >9.03 ng/mL or >11.41 when a formula with summation of EVLTLC and CyAC2 is used
Incidence of nephrogenic systemic fibrosis after administration of gadoteric acid in patients on renal replacement treatment
No full textNephrogenic system fibrosis (NSF) is a rare complication detected in patients with renal insufficiency exposed to gadolinium-based contrast agents (GBCAs). The aim of our study is to evaluate the prevalence of NSF in a cohort of patients on renal replacement treatment who underwent GBCA-enhanced magnetic resonance imaging (MRI)
Hardware design automation of convolutional neural networks
No full textConvolutional Neural Networks (CNNs) are a variation of feed-forward Neural Networks inspired by the biological process in the visual cortex of animals. The interest in this supervised learning algorithm has rapidly grown in many fields like image and video recognition and natural language processing. Nowadays they have become the state of the art in various applications like mobile robot vision, video surveillance and Big Data analytics. The specific computation pattern of CNNs results to be highly suitable for hardware acceleration, in fact different types of accelerators have been proposed based on GPU, Field Programmable Gate Array (FPGA) and ASIC. In particular, in the embedded systems context, due to real time and power consumption challenges, it is crucial to find the right tradeoff between performance, energy efficiency, fast development round and cost. This work proposes a framework meant as a tool for the user to accelerate and simplify the design and the implementation of CNNs on FPGAS by leveraging High Level Synthesis, still providing a certain level of customization of the hardware design
Esposizione a particolato atmosferico indoor e variazioni stagionali possono influire sul microbioma nasale di soggetti sani? Risultati preliminari.
No full textAndrea Camilleri y las reescrituras en la ficción televisiva del comisario Montalbano.
No full textCopyright del editor y de los autores.Es gracias al enorme éxito editorial de Andrea Camilleri (Porto Empedocle, 1925 – Roma, 2019) y, más concretamente, al ingente patrimonio literario que nos ha dejado, si nos encontramos hoy ante un mito cuya universalidad, celebrada por lectores, críticos, y también por buena parte del mundo académico, sigue todavía despertando enorme interés a tres años de su muerte. Camilleri ha escrito alrededor de 132 novelas, traducidas a 38 idiomas en todo el mundo : muchas de sus obras han sido objeto de adaptaciones fílmicas cuyo resultado trataremos analizar en este trabajo. Tuvimos la fortuna de conocer a Camilleri y poder trabajar sobre su obra: intercambiamos experiencias y en muchas ocasiones dialogamos también acerca de las trasposiciones fílmicas de su obra, de las cuales estaba sumamente satisfecho.
Para entender los motivos del óptimo resultado obtenido en el trasvase fílmico de la serie literaria de Camilleri, intentaremos descifrar los principales factores que han consolidado sobretodo el éxito de la serie de El comisario Montalbano, verdadera “joya” del panorama televisivo italiano (Solazzo, 2016) . El autor participó activamente en la puesta en escena de sus novelas, conforme a sus ideas creativas, limando las asperezas del texto, poniendo un toque personal en la elección del entorno geográfico y cultural. Lo que aquí quisiéramos destacar es principalmente la fidelidad de la serie fílmica al texto narrativo: en ella se refleja no sólo el buen conseguimiento del “trasvase” del guion literario, sino también el ajuste perfecto al estilo del autor
An exclusionary screening method based on 3D morphometric features to sort commingled atlases and axes
No full textIn forensic commingled contexts, when the disarticulation occurs uniquely at the atlantoaxial joint, the correct match of atlas and axis may lead to the desirable assembly of the entire body. Notwithstanding the importance of this joint in such scenarios, no study has so far explored three-dimensional (3D) methodologies to match these two adjoining bones. In the present study, we investigated the potential of re-associating atlas and axis through 3D-3D superimposition by testing their articular surfaces congruency in terms of point-to-point distance (Root Mean Square, RMS). We analysed vertebrae either from the same individual (match) and from different individuals (mismatch). The RMS distance values were assessed for both groups (matches and mismatches) and a threshold value was determined to discriminate matches with a sensitivity of 100%. The atlas and the corresponding axis from 41 documented skeletons (18 males and 23 females), in addition to unpaired elements (the atlas or the axis) from 5 individuals, were superimposed, resulting in 41 matches and 1851 mismatches (joining and non-joining elements). No sex-related significant differences were found in matches and mismatches (p = 0.270 and p = 0.210, respectively), allowing to pool together the two sexes in each group. RMS values ranged between 0.41 to 0.77 mm for matches and between 0.37 and 2.18 mm for mismatches. Significant differences were found comparing the two groups (p < 0.001) and the highest RMS of matches (0.77 mm) was used as the discriminative value that provided a sensitivity of 100% and a specificity of 41%. In conclusion, the 3D-3D superimposition of the atlanto-axial articular facets cannot be considered as a re-association method per se, but rather as a screening one. However, further research on the validation of the 3D approach and on its application to other joints might provide clues to the complex topic of the reassociation of crucial adjoining bones
An automated design framework for FPGA-based hardware accelerators of convolutional neural networks
No full textLAUREA MAGISTRALELe Reti Neurali Convoluzionali (conosciute come Convolutional Neural
Networks) sono un particolare tipo di Rete Neurale Artificiale, il cui funzionamento è ispirato a cellule presenti nella corteccia visiva degli animali, e rappresenta oggi la miglior soluzione per il riconoscimento e la classificazione di immagini. Al giorno d’oggi, le cosiddette Convolutional Neural Networks (CNNs) e altri tipi di algoirtmi appartenenti alla branca del Deep Learning vengono ampiamente utilizzati in contesti come quelli
dell’analisi di big data e dei sistemi embedded smart, fornendo delle tecnologie personalizzate attraverso servizi cloud-based e dispositivi come ad esempio smartphones, e smart watches. In questo tipo di applicazioni, l’enorme mole di dati da processare e i vincoli
di consumo energetico rendono cruciale l’individuazione di soluzioni che siano sia veloci che efficienti da un punto di vista energetico. In particolare, lo specifico flusso di calcolo di una CNN rende questo tipo di algoritmi estremamente adatti per essere accelerati in dispositivi hardware dedicati.
Infatti, molti acceleratori hardware basati su Graphics Processing Units (GPUs) ,Field-Programmable Gate Arrays (FPGAs) ed Application-Specific Integrated Circuits (ASICs) sono stati proposti a questo scopo. Tra questi, le FPGA sono in grado di fornire un giusto compromesso tra flessibilità, performance e consumo energetico. Tuttavia, il design e l’implementazione di un acceleratore per una CNN su questo tipo di dispositivi potrebbe risultare sia complesso che oneroso in termini di tempo di sviluppo, specialmente per sviluppatori con poca esperienza di progettazione hardware.
Per questi motivi, il lavoro proposto in questa tesi propone un framework in grado di generare automaticamente un’implementazione hardware di CNN su FPGA attraverso strumenti di High Level Synthesis (HLS). Il flusso di lavoro del framework parte da una descrizione ad alto livello della rete, integrandosi con il framework di Machine Learning TensorFlow per l’addestramento e una libreria C++ sviluppata internamente per l’implementazione finale.Convolutional Neural Networks are a particular type of Artificial Neural Networks (ANNs) inspired by the biological processes in the primary visual cortex of animals, and represent the state of the art in image recog-
nition and classification. Nowadays, Convolutional Neural Networks
(CNNs) and other Deep Learning algorithms have been extensively adopted
in contexts such as big data analysis and smart embedded systems, providing customized technologies through cloud-based services and personalized devices.
As regards this type of applications, the huge amount of data to be processed and power constraints require to find techniques to build fast and energy efficient solutions. In particular, the dataflow pattern of CNN algo-
rithm make them highly suitable for hardware acceleration. In fact, many hardware accelerators have been proposed based on Graphics Processing
Units (GPUs) , Field-Programmable Gate Arrays (FPGAs) , Application-Specific Integrated Circuits (ASICs) . Among them, FPGAs are able to make a proper tradeoff between flexibility, performance and power consumption. However, the design and the implementation of a CNN accelerator on such devices may result complex and time consuming, especially for developers that are not experienced in hardware design.
For these reasons, the work presented in this thesis proposes a framework to automatically generate a hardware implementation of CNNs on
FPGAs though High Level Synthesis (HLS) tools. The working flow of the framework starts from an high level description of the network, integrating TensorFlow for training and an internally developed C++ library for the final implementation
Proposta di traduzione di un libro di poesia illustrato per l’infanzia: Fleco de nube di Fabiana Margolis
No full textL'elaborato nasce da un interesse personale per la poesia per bambini e la sua traduzione. Si è quindi scelto, come oggetto di analisi e proposta di traduzione, la raccolta poetica illustrata "Fleco de nube" dell'autrice argentina Fabiana Margolis. Il volume, insignito del XV Premio Orihuela nel 2022, è un invito a scoprire la bellezza delle piccole cose e della Natura attraverso lo sguardo attento dell'infanzia. Sono presenti tre capitoli: nel primo si riporta l'analisi del testo di partenza, funzionale alla proposta di traduzione dei 15 componimenti, che trova spazio nel secondo capitolo, mentre nel terzo si offre il commento alla traduzione
Locally Advanced Cervical Cancer: Neoadjuvant Treatment versus Standard Radio-Chemotherapy—An Updated Meta-Analysis
No full text- …
